The scaleup system reaches five percent of founders. The other ninety-five percent are described as failures.

They are not failures. They are the unbuilt-for.

The bandwidth problem

7.5 million scalable startups in the world. Fewer than five in a hundred ever reach what we typically regard as scale: a $50M valuation or a successful Series A. The rate has not moved in decades. It is treated as a natural law of entrepreneurship, like gravity.

It is not. It is bandwidth.

Y Combinator has backed around 5,000 companies in twenty years. The Winter 2024 cohort took 260 from more than 27,000 applications. An acceptance rate of 0.96%. Seven in ten of its founders are based in the US; 99% of its returns have come from companies based there.

The UK has more than 500 accelerator programmes. Only 57% are still active. A third have closed. One in ten sits in limbo. Outside London, more than half have shut down.

These are not failures of effort. The people inside those programmes are doing serious work. They are failures of capacity. The programmes that exist were never going to scale. The arithmetic was always against it.

You cannot accelerator your way to a global solution. You cannot incubator your way to 7.5 million.

What the programmes actually do

The bandwidth problem is painful because the programmes work.

Wharton's Assenova and Amit looked at 8,580 startups across 408 accelerators in 176 countries. Accelerated companies were 3.4 percentage points more likely to raise venture capital. Among those that did raise, the average was $1.8M higher in the first year post-graduation. Y Combinator, the strongest case, produces a 45% Series A rate against a baseline of 33%. By 2015, around a third of US Series A rounds went to accelerator graduates — up from 10% in 2008.

Catalini, Guzman and Stern's MIT Sloan analysis of ten million firms: VC participation multiplies the probability of high-growth outcomes by five.

These are some of the largest causal findings in any business intervention literature.

The effects are real. The reach is the problem.

The problem is who gets them, not whether they work. Brookings reviewed nearly 700 US programmes that called themselves accelerators. Fewer than one in three met the formal definition. The rest were workspaces, co-working setups, business support wearing the label. The real effects concentrate in the strongest programmes; the long tail is undifferentiated noise.

So: seven million-plus scalable startups. A support system with proven effects, reaching less than 5% of them. A narrative that calls the missing 95% natural selection.

The mythology has done its work

The ecosystem tells itself a story. The founders who scale are exceptional. The founders who don't are not. The system's job is to back the exceptional. The evidence for the story is the power law of venture returns. The evidence is real.

Correlation Ventures, more than 21,000 venture-backed startups, 2004 to 2013. 65% return less than capital. 4% return more than 10x. The top 0.4% return more than 50x. Horsley Bridge, 7,000 startups between 1985 and 2014: 5% of capital deployed produced 60% of returns. Peter Thiel, Zero to One: the best investment in a successful fund will roughly equal or outperform the rest of the fund combined.

The ecosystem cites this as a justification of its model. It is an indictment of it.

If elite-programme returns themselves depend on one or two outliers per cohort, concentrating support on the top five percent is structurally fragile. The system's results are not produced by good selection. They are produced by accidents of distribution inside a small selected pool. You could not reliably reproduce them.

Cambridge Associates' US Venture Capital Index has delivered 14.3% CAGR over twenty-five years against the S&P 500's 7.4%. The headline is driven by a small number of top-quartile funds. The median VC fund does not outperform. Bottom-quartile funds lose money. Cambridge's own January 2026 commentary acknowledges that over the past five to ten years the index has struggled to keep up with the S&P 500 and the Nasdaq.

This is the inversion. A power law does not justify concentrating bets on the few. It justifies broadening the base. If outcomes are governed by hidden distributions you cannot identify in advance, the rational strategy is to give as many founders as possible the support that improves their position on the distribution. Not to filter ninety-five percent of them out at the door.

The messy middle, the companies that look average, sits inside that broader distribution. The system has spent forty years filtering it out, then turning around and admitting it cannot predict outcomes from the cohort it kept. The founders we should be most interested in are sitting outside the room.

Geography

Birmingham is not London. Manchester is not Cambridge. Marrakech is not Palo Alto. A founder in Birmingham is less likely to access venture-backed support than a founder in London. Not because of anything they have done. Because of where they happen to live.

Startup Genome's Global Startup Ecosystem Report 2024: Silicon Valley alone holds 59% of the value of the top five ecosystems. The top three — Silicon Valley, New York, London — have held station since 2020. The British Business Bank's 2025 Equity Tracker: London takes 61% of UK equity investment.

The further from those three a founder is, the smaller their effective ecosystem becomes.

This is not a moral failing. It is a structural one. Physical programmes cannot scale to global demand. They were never going to.

Small for many beats large for few

If bandwidth is the constraint, the arithmetic changes.

Lift the global scaleup rate by a single percentage point. Against a standing pool of 7.5 million scalable startups, that is 75,000 additional scaleups across the journey. A twenty percent increase in the world's scaleup output. Trillions in enterprise value at the threshold definition.

The Coutu Scale-Up Report of 2014 reached the same conclusion from the UK angle. Closing the UK scaleup gap would add tens of billions in turnover and hundreds of thousands of jobs.

Different methodology. Same answer.

The entire elite programme system reaches roughly 5% of founders. The other 95% scale at the base rate, with no structured support at all. A modest uplift across that base produces more aggregate scaleups than the elite system can reach in twenty years of operation.

A small improvement for many outweighs a large improvement for few.

That is not a slogan. It is what the arithmetic says. The power-law dynamics the ecosystem cites in its defence make it more true, not less.

What follows

Most of what gets called ecosystem innovation does not address the constraint. Co-working spaces wrapped as incubators. AI consulting tools without an evidence base. Founder communities behind a paywall. Some of it is real estate with a logo. Some of it is software with no underlying model. None of it brings a data-driven, 360-degree response to the core problem.

The constraint is solvable. Not by adding more accelerators. By building something that can reach the 95% at the marginal cost of software, with the rigour of the best programmes, and with no requirement that the founder be in the right city or know the right people.

That is why we built Pixie.

The 95% are not the failures. The system that reaches 5% of them and calls the job done is the failure.

Underlying research available on request.